AI Threat and Vulnerability Management AAISM Practice Questions (Page 8)

Questions 36–38

1. 36foundation · easy

   What is the primary goal of an adversarial example attack?

   • ATo steal the model's architecture by querying it
   • BTo cause the model to make a wrong prediction with a subtly modified input
   • CTo overload the model with requests to cause downtime
   • DTo corrupt the training data to introduce bias
   Reveal Solution Discussion Select an answer first
2. 37application · medium

   An e-commerce company discovers that an attacker has inserted a backdoor into its product recommendation model. The backdoor triggers when a specific pattern appears in user browsing history, causing the model to recommend competitor products. The company needs to remediate this quickly. Which action is most appropriate as the first step?

   • AIncrease monitoring to detect when the backdoor is triggered.
   • BAdd input sanitization to filter out the trigger pattern.
   • CRetrain the model from scratch using the original training data.
   • DRoll back the model to the last known good version and verify its integrity.
   Reveal Solution Discussion Select an answer first
3. 38application · medium

   A security analyst is investigating a potential evasion attack on a spam filter. The filter is misclassifying some emails as legitimate, and the analyst suspects that the attacker is crafting emails that bypass the filter. Which technique is most effective for detecting this type of attack?

   • AMonitor the model's confidence scores and flag emails with unusually high confidence that are later reported as spam by users.
   • BCompare the content of emails to known spam patterns using a signature-based system.
   • CBlock all emails from unknown senders.
   • DIncrease the frequency of model retraining to keep up with new attack patterns.
   Reveal Solution Discussion Select an answer first